Report on the Panel: "How Can Computer Architecture Researchers Avoid Becoming the Society for Irreproducible Results?" The First High-Performance Computer Architecture Symposium, Raleigh NC, Tuesday 24th January. Trevor Mudge, The University of Michigan, Ann Arbor The High-Performance Computer Architecture Conference, although young even among computer conferences (the second is scheduled for February 96), is already recognized as one of the leading forums for computer architecture research. At the first HPCA earlier this year I organized a panel with the above fl0e. It was purposely tongue-in-check, however, the intent was to provoke discussion on a point that distinguishes much of the experimental work in Computer Architecture from that of traditional experimental sciences: many results that are pubfished are difficult or impossible to confirm. As a starting point for the panelists I suggested the following reasons for this irreproducibility: 1. The authors: they give incomplete information about their experimental procedures. 2. Lack of access to the relevant experimental setup. 3. The reward system: their is no kudos for validating experiments (unlike in the bio sciences) m this means no tenure or research funding for confma'ting experiments. We were very fortunate to have a distinguished panel, who represented a wide range
/lp/association-for-computing-machinery/report-on-the-panel-how-can-computer-architecture-researchers-avoid-xKQnlqs8lp